perf: net_dropmonitor: Fix symbol-relative addresses
[linux-3.10.git] / kernel / ptrace.c
index 4f689cb..aed981a 100644 (file)
@@ -8,7 +8,7 @@
  */
 
 #include <linux/capability.h>
-#include <linux/module.h>
+#include <linux/export.h>
 #include <linux/sched.h>
 #include <linux/errno.h>
 #include <linux/mm.h>
 #include <linux/ptrace.h>
 #include <linux/security.h>
 #include <linux/signal.h>
+#include <linux/uio.h>
 #include <linux/audit.h>
 #include <linux/pid_namespace.h>
 #include <linux/syscalls.h>
 #include <linux/uaccess.h>
 #include <linux/regset.h>
 #include <linux/hw_breakpoint.h>
+#include <linux/cn_proc.h>
+#include <linux/compat.h>
+
 
+static int ptrace_trapping_sleep_fn(void *flags)
+{
+       schedule();
+       return 0;
+}
 
 /*
  * ptrace a task: make the debugger its new parent and
@@ -77,13 +86,31 @@ void __ptrace_unlink(struct task_struct *child)
        spin_lock(&child->sighand->siglock);
 
        /*
-        * Reinstate GROUP_STOP_PENDING if group stop is in effect and
+        * Clear all pending traps and TRAPPING.  TRAPPING should be
+        * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
+        */
+       task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
+       task_clear_jobctl_trapping(child);
+
+       /*
+        * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
         * @child isn't dead.
         */
        if (!(child->flags & PF_EXITING) &&
            (child->signal->flags & SIGNAL_STOP_STOPPED ||
-            child->signal->group_stop_count))
-               child->group_stop |= GROUP_STOP_PENDING;
+            child->signal->group_stop_count)) {
+               child->jobctl |= JOBCTL_STOP_PENDING;
+
+               /*
+                * This is only possible if this thread was cloned by the
+                * traced task running in the stopped group, set the signal
+                * for the future reports.
+                * FIXME: we should change ptrace_init_task() to handle this
+                * case.
+                */
+               if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
+                       child->jobctl |= SIGSTOP;
+       }
 
        /*
         * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
@@ -91,16 +118,64 @@ void __ptrace_unlink(struct task_struct *child)
         * is in TASK_TRACED; otherwise, we might unduly disrupt
         * TASK_KILLABLE sleeps.
         */
-       if (child->group_stop & GROUP_STOP_PENDING || task_is_traced(child))
-               signal_wake_up(child, task_is_traced(child));
+       if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
+               ptrace_signal_wake_up(child, true);
 
        spin_unlock(&child->sighand->siglock);
 }
 
-/*
- * Check that we have indeed attached to the thing..
+/* Ensure that nothing can wake it up, even SIGKILL */
+static bool ptrace_freeze_traced(struct task_struct *task)
+{
+       bool ret = false;
+
+       /* Lockless, nobody but us can set this flag */
+       if (task->jobctl & JOBCTL_LISTENING)
+               return ret;
+
+       spin_lock_irq(&task->sighand->siglock);
+       if (task_is_traced(task) && !__fatal_signal_pending(task)) {
+               task->state = __TASK_TRACED;
+               ret = true;
+       }
+       spin_unlock_irq(&task->sighand->siglock);
+
+       return ret;
+}
+
+static void ptrace_unfreeze_traced(struct task_struct *task)
+{
+       if (task->state != __TASK_TRACED)
+               return;
+
+       WARN_ON(!task->ptrace || task->parent != current);
+
+       spin_lock_irq(&task->sighand->siglock);
+       if (__fatal_signal_pending(task))
+               wake_up_state(task, __TASK_TRACED);
+       else
+               task->state = TASK_TRACED;
+       spin_unlock_irq(&task->sighand->siglock);
+}
+
+/**
+ * ptrace_check_attach - check whether ptracee is ready for ptrace operation
+ * @child: ptracee to check for
+ * @ignore_state: don't check whether @child is currently %TASK_TRACED
+ *
+ * Check whether @child is being ptraced by %current and ready for further
+ * ptrace operations.  If @ignore_state is %false, @child also should be in
+ * %TASK_TRACED state and on return the child is guaranteed to be traced
+ * and not executing.  If @ignore_state is %true, @child can be in any
+ * state.
+ *
+ * CONTEXT:
+ * Grabs and releases tasklist_lock and @child->sighand->siglock.
+ *
+ * RETURNS:
+ * 0 on success, -ESRCH if %child is not ready.
  */
-int ptrace_check_attach(struct task_struct *child, int kill)
+static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
 {
        int ret = -ESRCH;
 
@@ -112,27 +187,42 @@ int ptrace_check_attach(struct task_struct *child, int kill)
         * be changed by us so it's not changing right after this.
         */
        read_lock(&tasklist_lock);
-       if ((child->ptrace & PT_PTRACED) && child->parent == current) {
+       if (child->ptrace && child->parent == current) {
+               WARN_ON(child->state == __TASK_TRACED);
                /*
                 * child->sighand can't be NULL, release_task()
                 * does ptrace_unlink() before __exit_signal().
                 */
-               spin_lock_irq(&child->sighand->siglock);
-               WARN_ON_ONCE(task_is_stopped(child));
-               if (task_is_traced(child) || kill)
+               if (ignore_state || ptrace_freeze_traced(child))
                        ret = 0;
-               spin_unlock_irq(&child->sighand->siglock);
        }
        read_unlock(&tasklist_lock);
 
-       if (!ret && !kill)
-               ret = wait_task_inactive(child, TASK_TRACED) ? 0 : -ESRCH;
+       if (!ret && !ignore_state) {
+               if (!wait_task_inactive(child, __TASK_TRACED)) {
+                       /*
+                        * This can only happen if may_ptrace_stop() fails and
+                        * ptrace_stop() changes ->state back to TASK_RUNNING,
+                        * so we should not worry about leaking __TASK_TRACED.
+                        */
+                       WARN_ON(child->state == __TASK_TRACED);
+                       ret = -ESRCH;
+               }
+       }
 
-       /* All systems go.. */
        return ret;
 }
 
-int __ptrace_may_access(struct task_struct *task, unsigned int mode)
+static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
+{
+       if (mode & PTRACE_MODE_NOAUDIT)
+               return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
+       else
+               return has_ns_capability(current, ns, CAP_SYS_PTRACE);
+}
+
+/* Returns 0 on success, -errno on denial. */
+static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 {
        const struct cred *cred = current_cred(), *tcred;
 
@@ -150,15 +240,14 @@ int __ptrace_may_access(struct task_struct *task, unsigned int mode)
                return 0;
        rcu_read_lock();
        tcred = __task_cred(task);
-       if (cred->user->user_ns == tcred->user->user_ns &&
-           (cred->uid == tcred->euid &&
-            cred->uid == tcred->suid &&
-            cred->uid == tcred->uid  &&
-            cred->gid == tcred->egid &&
-            cred->gid == tcred->sgid &&
-            cred->gid == tcred->gid))
+       if (uid_eq(cred->uid, tcred->euid) &&
+           uid_eq(cred->uid, tcred->suid) &&
+           uid_eq(cred->uid, tcred->uid)  &&
+           gid_eq(cred->gid, tcred->egid) &&
+           gid_eq(cred->gid, tcred->sgid) &&
+           gid_eq(cred->gid, tcred->gid))
                goto ok;
-       if (ns_capable(tcred->user->user_ns, CAP_SYS_PTRACE))
+       if (ptrace_has_cap(tcred->user_ns, mode))
                goto ok;
        rcu_read_unlock();
        return -EPERM;
@@ -167,8 +256,12 @@ ok:
        smp_rmb();
        if (task->mm)
                dumpable = get_dumpable(task->mm);
-       if (!dumpable && !task_ns_capable(task, CAP_SYS_PTRACE))
+       rcu_read_lock();
+       if (!dumpable && !ptrace_has_cap(__task_cred(task)->user_ns, mode)) {
+               rcu_read_unlock();
                return -EPERM;
+       }
+       rcu_read_unlock();
 
        return security_ptrace_access_check(task, mode);
 }
@@ -182,10 +275,24 @@ bool ptrace_may_access(struct task_struct *task, unsigned int mode)
        return !err;
 }
 
-static int ptrace_attach(struct task_struct *task)
+static int ptrace_attach(struct task_struct *task, long request,
+                        unsigned long addr,
+                        unsigned long flags)
 {
+       bool seize = (request == PTRACE_SEIZE);
        int retval;
 
+       retval = -EIO;
+       if (seize) {
+               if (addr != 0)
+                       goto out;
+               if (flags & ~(unsigned long)PTRACE_O_MASK)
+                       goto out;
+               flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
+       } else {
+               flags = PT_PTRACED;
+       }
+
        audit_ptrace(task);
 
        retval = -EPERM;
@@ -196,7 +303,7 @@ static int ptrace_attach(struct task_struct *task)
 
        /*
         * Protect exec's credential calculations against our interference;
-        * interference; SUID, SGID and LSM creds get determined differently
+        * SUID, SGID and LSM creds get determined differently
         * under ptrace.
         */
        retval = -ERESTARTNOINTR;
@@ -216,17 +323,24 @@ static int ptrace_attach(struct task_struct *task)
        if (task->ptrace)
                goto unlock_tasklist;
 
-       task->ptrace = PT_PTRACED;
-       if (task_ns_capable(task, CAP_SYS_PTRACE))
-               task->ptrace |= PT_PTRACE_CAP;
+       if (seize)
+               flags |= PT_SEIZED;
+       rcu_read_lock();
+       if (ns_capable(__task_cred(task)->user_ns, CAP_SYS_PTRACE))
+               flags |= PT_PTRACE_CAP;
+       rcu_read_unlock();
+       task->ptrace = flags;
 
        __ptrace_link(task, current);
-       send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
+
+       /* SEIZE doesn't trap tracee on attach */
+       if (!seize)
+               send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
 
        spin_lock(&task->sighand->siglock);
 
        /*
-        * If the task is already STOPPED, set GROUP_STOP_PENDING and
+        * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
         * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
         * will be cleared if the child completes the transition or any
         * event which clears the group stop states happens.  We'll wait
@@ -242,10 +356,9 @@ static int ptrace_attach(struct task_struct *task)
         * The following task_is_stopped() test is safe as both transitions
         * in and out of STOPPED are protected by siglock.
         */
-       if (task_is_stopped(task)) {
-               task->group_stop |= GROUP_STOP_PENDING | GROUP_STOP_TRAPPING;
-               signal_wake_up(task, 1);
-       }
+       if (task_is_stopped(task) &&
+           task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
+               signal_wake_up_state(task, __TASK_STOPPED);
 
        spin_unlock(&task->sighand->siglock);
 
@@ -255,9 +368,12 @@ unlock_tasklist:
 unlock_creds:
        mutex_unlock(&task->signal->cred_guard_mutex);
 out:
-       if (!retval)
-               wait_event(current->signal->wait_chldexit,
-                          !(task->group_stop & GROUP_STOP_TRAPPING));
+       if (!retval) {
+               wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT,
+                           ptrace_trapping_sleep_fn, TASK_UNINTERRUPTIBLE);
+               proc_ptrace_connector(task, PTRACE_ATTACH);
+       }
+
        return retval;
 }
 
@@ -320,25 +436,27 @@ static int ignoring_children(struct sighand_struct *sigh)
  */
 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
 {
+       bool dead;
+
        __ptrace_unlink(p);
 
-       if (p->exit_state == EXIT_ZOMBIE) {
-               if (!task_detached(p) && thread_group_empty(p)) {
-                       if (!same_thread_group(p->real_parent, tracer))
-                               do_notify_parent(p, p->exit_signal);
-                       else if (ignoring_children(tracer->sighand)) {
-                               __wake_up_parent(p, tracer);
-                               p->exit_signal = -1;
-                       }
-               }
-               if (task_detached(p)) {
-                       /* Mark it as in the process of being reaped. */
-                       p->exit_state = EXIT_DEAD;
-                       return true;
+       if (p->exit_state != EXIT_ZOMBIE)
+               return false;
+
+       dead = !thread_group_leader(p);
+
+       if (!dead && thread_group_empty(p)) {
+               if (!same_thread_group(p->real_parent, tracer))
+                       dead = do_notify_parent(p, p->exit_signal);
+               else if (ignoring_children(tracer->sighand)) {
+                       __wake_up_parent(p, tracer);
+                       dead = true;
                }
        }
-
-       return false;
+       /* Mark it as in the process of being reaped. */
+       if (dead)
+               p->exit_state = EXIT_DEAD;
+       return dead;
 }
 
 static int ptrace_detach(struct task_struct *child, unsigned int data)
@@ -363,6 +481,7 @@ static int ptrace_detach(struct task_struct *child, unsigned int data)
        }
        write_unlock_irq(&tasklist_lock);
 
+       proc_ptrace_connector(child, PTRACE_DETACH);
        if (unlikely(dead))
                release_task(child);
 
@@ -385,6 +504,9 @@ void exit_ptrace(struct task_struct *tracer)
                return;
 
        list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
+               if (unlikely(p->ptrace & PT_EXITKILL))
+                       send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
+
                if (__ptrace_detach(tracer, p))
                        list_add(&p->ptrace_entry, &ptrace_dead);
        }
@@ -452,30 +574,18 @@ int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long ds
 
 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 {
-       child->ptrace &= ~PT_TRACE_MASK;
-
-       if (data & PTRACE_O_TRACESYSGOOD)
-               child->ptrace |= PT_TRACESYSGOOD;
-
-       if (data & PTRACE_O_TRACEFORK)
-               child->ptrace |= PT_TRACE_FORK;
-
-       if (data & PTRACE_O_TRACEVFORK)
-               child->ptrace |= PT_TRACE_VFORK;
+       unsigned flags;
 
-       if (data & PTRACE_O_TRACECLONE)
-               child->ptrace |= PT_TRACE_CLONE;
-
-       if (data & PTRACE_O_TRACEEXEC)
-               child->ptrace |= PT_TRACE_EXEC;
-
-       if (data & PTRACE_O_TRACEVFORKDONE)
-               child->ptrace |= PT_TRACE_VFORK_DONE;
+       if (data & ~(unsigned long)PTRACE_O_MASK)
+               return -EINVAL;
 
-       if (data & PTRACE_O_TRACEEXIT)
-               child->ptrace |= PT_TRACE_EXIT;
+       /* Avoid intermediate state when all opts are cleared */
+       flags = child->ptrace;
+       flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
+       flags |= (data << PT_OPT_FLAG_SHIFT);
+       child->ptrace = flags;
 
-       return (data & ~PTRACE_O_MASK) ? -EINVAL : 0;
+       return 0;
 }
 
 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
@@ -510,6 +620,81 @@ static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
        return error;
 }
 
+static int ptrace_peek_siginfo(struct task_struct *child,
+                               unsigned long addr,
+                               unsigned long data)
+{
+       struct ptrace_peeksiginfo_args arg;
+       struct sigpending *pending;
+       struct sigqueue *q;
+       int ret, i;
+
+       ret = copy_from_user(&arg, (void __user *) addr,
+                               sizeof(struct ptrace_peeksiginfo_args));
+       if (ret)
+               return -EFAULT;
+
+       if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
+               return -EINVAL; /* unknown flags */
+
+       if (arg.nr < 0)
+               return -EINVAL;
+
+       if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
+               pending = &child->signal->shared_pending;
+       else
+               pending = &child->pending;
+
+       for (i = 0; i < arg.nr; ) {
+               siginfo_t info;
+               s32 off = arg.off + i;
+
+               spin_lock_irq(&child->sighand->siglock);
+               list_for_each_entry(q, &pending->list, list) {
+                       if (!off--) {
+                               copy_siginfo(&info, &q->info);
+                               break;
+                       }
+               }
+               spin_unlock_irq(&child->sighand->siglock);
+
+               if (off >= 0) /* beyond the end of the list */
+                       break;
+
+#ifdef CONFIG_COMPAT
+               if (unlikely(is_compat_task())) {
+                       compat_siginfo_t __user *uinfo = compat_ptr(data);
+
+                       ret = copy_siginfo_to_user32(uinfo, &info);
+                       ret |= __put_user(info.si_code, &uinfo->si_code);
+               } else
+#endif
+               {
+                       siginfo_t __user *uinfo = (siginfo_t __user *) data;
+
+                       ret = copy_siginfo_to_user(uinfo, &info);
+                       ret |= __put_user(info.si_code, &uinfo->si_code);
+               }
+
+               if (ret) {
+                       ret = -EFAULT;
+                       break;
+               }
+
+               data += sizeof(siginfo_t);
+               i++;
+
+               if (signal_pending(current))
+                       break;
+
+               cond_resched();
+       }
+
+       if (i > 0)
+               return i;
+
+       return ret;
+}
 
 #ifdef PTRACE_SINGLESTEP
 #define is_singlestep(request)         ((request) == PTRACE_SINGLESTEP)
@@ -604,15 +789,23 @@ static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
                                             kiov->iov_len, kiov->iov_base);
 }
 
+/*
+ * This is declared in linux/regset.h and defined in machine-dependent
+ * code.  We put the export here, near the primary machine-neutral use,
+ * to ensure no machine forgets it.
+ */
+EXPORT_SYMBOL_GPL(task_user_regset_view);
 #endif
 
 int ptrace_request(struct task_struct *child, long request,
                   unsigned long addr, unsigned long data)
 {
+       bool seized = child->ptrace & PT_SEIZED;
        int ret = -EIO;
-       siginfo_t siginfo;
+       siginfo_t siginfo, *si;
        void __user *datavp = (void __user *) data;
        unsigned long __user *datalp = datavp;
+       unsigned long flags;
 
        switch (request) {
        case PTRACE_PEEKTEXT:
@@ -632,6 +825,10 @@ int ptrace_request(struct task_struct *child, long request,
                ret = put_user(child->ptrace_message, datalp);
                break;
 
+       case PTRACE_PEEKSIGINFO:
+               ret = ptrace_peek_siginfo(child, addr, data);
+               break;
+
        case PTRACE_GETSIGINFO:
                ret = ptrace_getsiginfo(child, &siginfo);
                if (!ret)
@@ -645,6 +842,59 @@ int ptrace_request(struct task_struct *child, long request,
                        ret = ptrace_setsiginfo(child, &siginfo);
                break;
 
+       case PTRACE_INTERRUPT:
+               /*
+                * Stop tracee without any side-effect on signal or job
+                * control.  At least one trap is guaranteed to happen
+                * after this request.  If @child is already trapped, the
+                * current trap is not disturbed and another trap will
+                * happen after the current trap is ended with PTRACE_CONT.
+                *
+                * The actual trap might not be PTRACE_EVENT_STOP trap but
+                * the pending condition is cleared regardless.
+                */
+               if (unlikely(!seized || !lock_task_sighand(child, &flags)))
+                       break;
+
+               /*
+                * INTERRUPT doesn't disturb existing trap sans one
+                * exception.  If ptracer issued LISTEN for the current
+                * STOP, this INTERRUPT should clear LISTEN and re-trap
+                * tracee into STOP.
+                */
+               if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
+                       ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
+
+               unlock_task_sighand(child, &flags);
+               ret = 0;
+               break;
+
+       case PTRACE_LISTEN:
+               /*
+                * Listen for events.  Tracee must be in STOP.  It's not
+                * resumed per-se but is not considered to be in TRACED by
+                * wait(2) or ptrace(2).  If an async event (e.g. group
+                * stop state change) happens, tracee will enter STOP trap
+                * again.  Alternatively, ptracer can issue INTERRUPT to
+                * finish listening and re-trap tracee into STOP.
+                */
+               if (unlikely(!seized || !lock_task_sighand(child, &flags)))
+                       break;
+
+               si = child->last_siginfo;
+               if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
+                       child->jobctl |= JOBCTL_LISTENING;
+                       /*
+                        * If NOTIFY is set, it means event happened between
+                        * start of this trap and now.  Trigger re-trap.
+                        */
+                       if (child->jobctl & JOBCTL_TRAP_NOTIFY)
+                               ptrace_signal_wake_up(child, true);
+                       ret = 0;
+               }
+               unlock_task_sighand(child, &flags);
+               break;
+
        case PTRACE_DETACH:      /* detach a process that was attached. */
                ret = ptrace_detach(child, data);
                break;
@@ -759,8 +1009,8 @@ SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
                goto out;
        }
 
-       if (request == PTRACE_ATTACH) {
-               ret = ptrace_attach(child);
+       if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
+               ret = ptrace_attach(child, request, addr, data);
                /*
                 * Some architectures need to do book-keeping after
                 * a ptrace attach.
@@ -770,11 +1020,14 @@ SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
                goto out_put_task_struct;
        }
 
-       ret = ptrace_check_attach(child, request == PTRACE_KILL);
+       ret = ptrace_check_attach(child, request == PTRACE_KILL ||
+                                 request == PTRACE_INTERRUPT);
        if (ret < 0)
                goto out_put_task_struct;
 
        ret = arch_ptrace(child, request, addr, data);
+       if (ret || request != PTRACE_DETACH)
+               ptrace_unfreeze_traced(child);
 
  out_put_task_struct:
        put_task_struct(child);
@@ -901,8 +1154,8 @@ asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
                goto out;
        }
 
-       if (request == PTRACE_ATTACH) {
-               ret = ptrace_attach(child);
+       if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
+               ret = ptrace_attach(child, request, addr, data);
                /*
                 * Some architectures need to do book-keeping after
                 * a ptrace attach.
@@ -912,9 +1165,13 @@ asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
                goto out_put_task_struct;
        }
 
-       ret = ptrace_check_attach(child, request == PTRACE_KILL);
-       if (!ret)
+       ret = ptrace_check_attach(child, request == PTRACE_KILL ||
+                                 request == PTRACE_INTERRUPT);
+       if (!ret) {
                ret = compat_arch_ptrace(child, request, addr, data);
+               if (ret || request != PTRACE_DETACH)
+                       ptrace_unfreeze_traced(child);
+       }
 
  out_put_task_struct:
        put_task_struct(child);